The items don’t seem concise and always clear. But seems like a good, inspiring resource for things to consider.
If it is expected that a method might fail, then it should fail, either by throwing an Exception or, if not - it should return a special case None/Null type object of the desired class (following the Null Object Pattern), not null itself.
I’ve never heard of evading null with a Null object. Seems like a bad idea to me. Maybe it could work in some language, but generally I would say prefer result typing. Introducing a result type wrapping or extending the result value type is complexity I would be very evasive to introduce if the language doesn’t already support result wrapper/state types.
We use null objects at work, and as another person said they are a safety feature. Here’s how they work: they are wrappers around another type. They provide the same interface as the wrapped type. They store one global instance of the wrapped type, default initialized, in a memory page marked read-only.
Here’s why they are considered a safety feature (note: most of this is specific to c++).
Suppose you have a collection, and you want to write a function that finds an item in the collection. Find can fail, of course. What do you return in that case? Reasonable options would be a null pointer, or
std::nullopt
. Having find return astd::optional
would be perfect, because that’s the exact use case for it. You either found the item or you did not.Now, the problem is that in most cases you don’t want to copy the item that was found in the collection when you return it, so you want to return a pointer or a reference. Well,
std::optional<T&>
is illegal. After all, an optional reference has the same semantics as a pointer, no? This means your find function cannot return an optional, it has to return a pointer with the special sentinel value ofnullptr
meaning “not found”.But returning
nullptr
is dangerous, because if you forget to check the return value and you accidentally dereference it you invoke undefined behavior which in this case will usually crash the program.Here’s where the null object comes in. You make find just return a reference. If the item is not found, you return a reference to the null object of the relevant type. Because the null object always exists, it’s not UB to access it. And because it is default initialized, trying to get a value from it will just give you the default value for that data member.
Basically it’s a pattern to avoid crashing if tou forget to check for nullptr
I’ve never heard of evading null with a Null object.
This is quite standard, and in fact it’s even a safety feature. C++ introduced nullptr defined as an instance of std::nullptr_t explicitly with this in mind.
https://en.cppreference.com/w/cpp/language/nullptr
This approach is also quite basic in monadic types.
with this in mind
With what in mind? Evading
NULL
?Languages that make use of references rather than pointers don’t have this Dualism. C# has nullable references and nullability analysis, and
null
as a keyword.What does your reasoning mean in that context?
Languages that make use of references rather than pointers don’t have this Dualism.
It’s not about references vs pointers. You could easily have a language that allowed “null references” (edit: too much C++; of course many languages allow null references, e.g. Javascript) or one that properly separated null pointers out in the type system.
I agree with your point though, using a special
Null
value is usually worse than usingOption
or similar. Andnullptr_t
doesn’t help with this at all.With what in mind? Evading NULL?
Depends on your perspective. It’s convenient to lean on type checking to avoid a whole class of bugs. You can see this either as avoiding NULL or use your type system to flag misuses.
Languages that make use of references rather than pointers don’t have this Dualism. C# has nullable references and nullability analysis, and null as a keyword.
C#'s
null
keyword matches the monadic approach I mentioned earlier. Nullable types work as aMaybe
monad. It’s the same concept shoehorned differently due to the different paths taken by these languages.as far as I know, C# don’t have proper ergonomic monadic bind as in F# (computation expression), Haskell (do expression), and Ocaml (let*), but I could be wrong.
Correct.
“Monadic type” has something like three meanings depending on context, and it’s not clear which one you mean. One of them is common in math, but not so common in programming, so probably not that. But neither “parametric types with a single argument” nor “types that encode a category-theoretic monad” have the property you say, as far as I know.
I imagine you’re probably referring to the latter, since the optional monad exists. That’s very different from returning null. The inhabitants of
Integer
in Java, for example, are the boxed machine ints andnull
. The inhabitants ofOptional[Integer]
(it won’t let me use angle brackets here) areOptional.of(i)
for each machine inti
,Optional.empty()
, andnull
.Optional.empty()
is not null and should not be called a “Null object.” It’s also not of typeInteger
, so you’re not even allowed to return it unless the function type explicitly says so. Writing such function types is pretty uncommon to do in java programs but it’s more normal in kotlin. In languages like Haskell, which don’t havenull
at all, this is idiomatic.I think you’re trying too hard to confuse yourself.
This might be educational: https://docs.oracle.com/javase/8/docs/api/java/util/Optional.html
There are issues that the
Optional
class alleviates that are common enough to be documented: https://www.jetbrains.com/help/inspectopedia/ConditionalCanBeOptional.html (more detail is available at places like https://github.com/JetBrains/intellij-community/blob/a2d32ec64ed0fb37c7cc97856aa94cce95b17ee5/java/java-impl/src/inspectionDescriptions/ConditionalCanBeOptional.html (I believe this information used to be visible with the “inspectopedia” URLs but I don’t see that today))On the other hand, it seems there are some features / situations that require
null
to be present: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Operators/Optional_chaining https://www.jetbrains.com/help/inspectopedia/OptionalToIf.html
This doesn’t seem overly useful.
It’s a list taken out of a bunch of books with no regard for how something can be the best path in one language and a smell in another language.
Look at this page for example: https://luzkan.github.io/smells/imperative-loops
It suggests using functional loop methods (
.map()
,.reduce()
,.filter()
) instead of using imperative loops (for
,for in
,for each
) but completely disregards the facts that imperative loops also have access to thebreak
,continue
, andreturn
keywords to improve performance.For example: If I have an unsorted list of 1000 cars which includes a whole bunch of information per car (e.g. color, year manufactured, etc…), and I want to know if there were any cars were manufactured before the year 1980, I can run an imperative loop through the list and early return true if I find one, and only returning false if I haven’t found one by the end of the list.
If the third car was made in 1977, then I have only iterated through 3 cars to find my answer.
But if I were to try this with only functional loops, I would have to iterate through all 1000 cars before I had my answer.
A website with blind rules like this is going to lead to worse code.
That’s a pretty bad example since most functional frameworks include an any or some function that returns early.
…what? At least with Java Streams or Kotlin Sequences, they absolutely abort early with something like
.filter().first()
.Same in Python, Rust, Haskell and probably many others.
But apparently JS does work that way, that is its
filter
always iterates over everything and returns a new array and not some iterator object.The old methods on Array will eagerly evaluate all elements. But JS has a new Iterator type with methods that works lazily instead.
Ya, streams may seem tedious (why do I have to call stream and collect?), but it’s like that for performance (and probably backwards compatibility).
If writing readable code is not peformant, then the language implementation needs to be fixed.
Honestly, it is much more code to use loop with non-local control like break, continue etc. (variable initialization, append, variable mutation in loops…) than just calling a collect function (which I assume just means to_list). In the above example, in most programming language I know, you don’t even need to collect the result into a list.
Not to mention, large loops with non-local control is a breeding ground for spegatti code. Because you no longer have a consistent exit point to the loop, thus making the semantics hard o reason about.
In many languages, there are type class / trait / interfaces (whatever you want to call them) that allows lazy structures to share the same API as strict ones.
Yeah, in Java calling
first()
on a stream is the same as an early return in a for-loop, where for each element all of the previous stream operations are applied first.So the stream operation
cars.stream() .filter(c -> c.year() < 1977) .first()
is equivalent to doing the following imperatively
for (var car : cars) { if (car.year() < 1977) return car; }
Not to mention Kotlin actually supports non-local returns in lambdas under specific circumstances, which allows for even more circumstances to be expressed with functional chaining.
These are not quite equivalent. In terms of short-circuiting yeah they both short-circuit when they get the value. But the latter is returning from the current function and the former is not. If you add a return to that first example then they are equivalent. But then cannot be used in line. Which is a nice advantage to the former - it can be used inline with less faff as you can just assign the return to a value. The latter needs you to declare a variable, assign it and break from the loop in the if.
Personally I quite like how the former requires less modification to work in different contexts and find it nicer to read. Though not all logic is easier to read with a stream, sometimes a good old for loop makes the code more readable. Use which ever helps you best at each point. Never blindly apply some pattern to every situation.
Well yes, I was simplifying because I wanted to address the main (incorrect) criticism by @spartanatreyu@programming.dev. I agree with your comment
Also, Effective Java specifically says to use streams judiciously and prefer traditional for loops in general.
This is like SCPs but for devs
You’re thinking of CVEs.